Yi-1.5-34B-32K

MERA Created at 20.09.2024 11:39
0.48
The overall result
182
Place in the rating
Weak tasks:
194
RWSD
296
PARus
299
RCB
232
ruEthics
309
MultiQ
313
ruWorldTree
304
ruOpenBookQA
332
CheGeKa
267
ruMMLU
282
ruHateSpeech
334
ruDetox
293
ruHHH
297
ruTiE
72
ruHumanEval
245
USE
120
MathLogicQA
118
ruMultiAr
144
SimpleAr
446
LCS
221
BPS
102
ruModAr
297
MaMuRAMu
69
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.078 Accuracy
RCB 0.516 / 0.48 Accuracy F1 macro
USE 0.162 Grade norm
RWSD 0.562 Accuracy
PARus 0.758 Accuracy
ruTiE 0.64 Accuracy
MultiQ 0.334 / 0.188 F1 Exact match
CheGeKa 0.068 / 0.048 F1 Exact match
ruModAr 0.688 Exact match
MaMuRAMu 0.609 Accuracy
ruMultiAr 0.34 Exact match
ruCodeEval 0.3 / 0.597 / 0.689 Pass@k
MathLogicQA 0.511 Accuracy
ruWorldTree 0.796 / 0.795 Accuracy F1 macro
ruOpenBookQA 0.708 / 0.706 Accuracy F1 macro

Evaluation on open tasks:

Go to the ratings by subcategory

The table will scroll to the left

Task name Result Metric
BPS 0.949 Accuracy
ruMMLU 0.546 Accuracy
SimpleAr 0.983 Exact match
ruHumanEval 0.282 / 0.543 / 0.64 Pass@k
ruHHH 0.635
ruHateSpeech 0.683
ruDetox 0.142
ruEthics
Correct God Ethical
Virtue 0.299 0.282 0.326
Law 0.308 0.276 0.328
Moral 0.324 0.308 0.359
Justice 0.263 0.241 0.298
Utilitarianism 0.257 0.302 0.265

Information about the submission

Mera version
v.1.2.0
Torch Version
2.3.1
The version of the codebase
aec92f8
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.43.3
The number of GPUs and their type
4 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Yi-1.5-34B-32K

Model size

34.4B

Model type:

Opened

Pretrain

Architecture description:

The Yi series models adopt the same model architecture as Llama but are NOT derivatives of Llama.

Description of the training:

Yi has independently created its own high-quality training datasets, efficient training pipelines, and robust training infrastructure entirely from the ground up.

Pretrain data:

Yi-1.5 is an upgraded version of Yi (which was trained on 3T multilingual corpus). It is continuously pre-trained on Yi with a high-quality corpus of 500B tokens and fine-tuned on 3M diverse fine-tuning samples.

License:

apache-2.0

Inference parameters

Generation Parameters:
simplear - do_sample=false;until=["\n"]; \nchegeka - do_sample=false;until=["\n"]; \nrudetox - do_sample=false;until=["\n"]; \nruhumaneval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6; \nrucodeeval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6; \nrumultiar - do_sample=false;until=["\n"]; \nuse - do_sample=false;until=["\n","."]; \nmultiq - do_sample=false;until=["\n"]; \nrumodar - do_sample=false;until=["\n"];

The size of the context:
simplear, bps, lcs, chegeka, mathlogicqa, parus, rcb, rudetox, ruhatespeech, rummlu, ruworldtree, ruethics, ruhhh, ruhumaneval, rucodeeval, mamuramu, rumultiar, ruopenbookqa, use, rwsd, multiq, rumodar, rutie - 32768

Ratings by subcategory

Metric: Grade Norm
Model, team 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 8_0 8_1 8_2 8_3 8_4
Yi-1.5-34B-32K
MERA
0.5 0.3 0.633 0.1 0 0.2 0.033 - 0 0.033 0 0 0.133 0 0.133 0.15 0 0.033 0.033 0 0 0.633 0.067 0.033 0.067 0.342 0.133 0.033 0.3 0.3 0.133
Model, team Honest Helpful Harmless
Yi-1.5-34B-32K
MERA
0.656 0.593 0.655
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Management Philosophy Prehistory Human aging Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Biology (college) Physics (college) Human Sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine learning Medical genetics Professional law PR Security studies Chemistry (школьная) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual_physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) European history Government and politics
Yi-1.5-34B-32K
MERA
0.363 0.44 0.572 0.744 0.621 0.687 0.66 0.588 0.565 0.529 0.395 0.524 0.43 0.694 0.59 0.618 0.66 0.507 0.4 0.565 0.391 0.561 0.37 0.445 0.366 0.52 0.434 0.602 0.698 0.33 0.73 0.793 0.589 0.717 0.562 0.573 0.44 0.603 0.417 0.532 0.652 0.456 0.566 0.631 0.635 0.556 0.642 0.467 0.411 0.49 0.54 0.688 0.613 0.622 0.84 0.697 0.617
Model, team SIM FL STA
Yi-1.5-34B-32K
MERA
0.752 0.608 0.371
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Managment Philosophy Pre-History Gerontology Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Bilology (college) Physics (college) Human sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine Learning Genetics Professional law PR Security Chemistry (college) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical Engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional Accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) Europe History Government and politics
Yi-1.5-34B-32K
MERA
0.4 0.525 0.5 0.611 0.487 0.655 0.483 0.526 0.365 0.585 0.744 0.583 0.467 0.667 0.526 0.593 0.701 0.578 0.456 0.667 0.509 0.746 0.8 0.479 0.644 0.47 0.641 0.544 0.789 0.689 0.8 0.718 0.652 0.667 0.47 0.661 0.644 0.733 0.596 0.554 0.612 0.429 0.8 0.889 0.879 0.778 0.603 0.864 0.708 0.789 0.756 0.507 0.747 0.584 0.721 0.427 0.722
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Yi-1.5-34B-32K
MERA
0.299 0.308 0.324 0.263 0.257
Model, team Virtue Law Moral Justice Utilitarianism
Yi-1.5-34B-32K
MERA
0.282 0.276 0.308 0.241 0.302
Model, team Virtue Law Moral Justice Utilitarianism
Yi-1.5-34B-32K
MERA
0.326 0.328 0.359 0.298 0.265
Model, team Women Men LGBT Nationalities Migrants Other
Yi-1.5-34B-32K
MERA
0.694 0.743 0.529 0.73 0 0.721